Sil Hamilton
I’m Sil. I’m a PhD student working with David Mimno and Matthew Wilkens at Cornell University in the Culture and Computation Lab. I'm also an AI Research Scientist at Epiq AI Labs. Here's my CV.
What do I do study? I look for where humans and language models disagree on the world. My published research draws on narrative theory to show how LLMs (fail to) capture cultural concepts in areas like jurisprudence, journalism, and storytelling. This helps us improve them.
What else? I've consulted for news organizations like NYT and AP on language models. I've also presented at the Nieman Foundation, CS50 at Harvard, and the Computer History Museum.
| 02/2026 | "Too Long, Didn't Model" was accepted by SIGHUM 2026 |
| 01/2026 | "NarraBench" was accepted by EACL 2026 |
| 10/2025 | Organizing the next NLP4DH, co-located with ACL 2026 |
| 07/2025 | Hired as an AI Research Scientist at Epiq AI Labs |
| 07/2025 | "The Zero Body Problem" was accepted by COLM 2025 |
| 05/2025 | Began a summer internship at Epiq AI Labs |
| 04/2025 | Attended NAACL 2025 and presented "A City of Millions" at NLP4DH |
| 04/2025 | NSERC awarded me a 3-year scholarship to pursue extracting knowledge from neural networks |
| 01/2025 | Spoke on structuring data for the digital humanities at Concordia University |
Simulation Papers
| Blind Judgement: Agent-Based Supreme Court Modelling with GPT. Simulating the Supreme Court with chatbots. |
| The COVID That Wasn’t: Counterfactual Journalism using GPT. Estimating priors on COVID-19 with GPT-2. |
Interpretability Papers
| The Zero Body Problem: Probing LLM Use of Sensory Language. Generated stories don't reliably evoke the senses. |
| Lost in Space: Finding the Right Tokens for Structured Output. Structured output harms model accuracy. |
| Detecting Mode Collapse in Language Models via Narration. RLHF causes GPT to write more generic stories. |
| Mrs. Dalloway Said She Would Segment the Chapters Herself. Finding chapter borders in story sentiment. |
Narrative Papers
| NarraBench: A Comprehensive Framework for Narrative Benchmarking. We need better benchmarks for capturing LLM creative performance. | |
| Too Long, Didn't Model: Decomposing LLM Long-Context Understanding With Novels. LLMs find processing novels challenging. | |
| A City of Millions: Mapping Literary Social Networks At Scale. Mapping literary social networks. | |
| MultiHATHI: A Complete Collection of Multilingual Prose Fiction in the HathiTrust Digital Library. Predicting fictionality with BERT. |
Software
| DocPlot. Private semantic search in the browser. |
| COVID-17. Showcases a counterfactual COVID narrative. |
| semantic-space. Generates thesauruses from latent space. |
| adsb-utils. Maps ADS-B packets with ncurses. |
| feature-space-explorer. Plots sentence embeddings in 3D. |
Courses
| Generative AI For Journalists. 350 students. |
| How to use ChatGPT and other generative AI tools in your newsrooms. 10k+ students and translated into Spanish & Portuguese. |
Workshops
| NLP4DH 2025. I served as session chair. |
| Workshop on AI & DH. Concordia University. |
| Run LLMs On Your Laptop. Media Party. |
| Workshop and Mixer on ChatGPT. Brown Institute for Media Innovation. |
Media
| Confessions of a Viral AI Writer. Wired. |
| Newsrooms Should Engage Responsibly with Artificial Intelligence, Say Journalists. Broadband Breakfast. |
| ChatGPT will not take our jobs, but it’s essential to experiment and learn about this technology, say experts in webinar. LatAm Journalism Review. |
| AI Can Reveal Hidden Biases in News Media. Futurity. |
| L’IA rendra-t-elle les médias plus objectifs? Le Devoir. |
| ChatGPT is less wowed by itself than we are. FT. |
| What AI-generated COVID news tells us that journalists don’t. McGill Channels. |
Speaking
| AI in journalism: Ethics, opportunities and pitfalls. ISOJ. |
| Artificial Intelligence in Modern Media Organizations. Baltic Media Forum. |
| How AI, Digital Health, & Technology Is Shaping Global Health & Humanitarian Relief. Kennedy Space Center. |
| Prometheus: Local News for Local AI. Media Party. |
| What Journalists Need To Know About ChatGPT and Other Tools. Knight Center, UTAustin. |
Lectures
| Open Lecture by Sil Hamilton. Vilnius University, Lithuania. |
| Digital Humanities 2.0. UNLV. |
| Mind the Gap: How Newsrooms Can Bridge the AI Moat. Computer History Museum. |
| GPT-4 - How does it work, and how do I build apps with it? CS50, Harvard. |
| The Present State: Generative AI. Nieman Foundation, Harvard. |
| Parallel Worlds: How To Generate Useful Data with AI. McGill University. |
| Superreaders: Recognizing The Value Of Literate Language Models. SLSA. |
Interviews
| The Future of CNF Writing, Editing and Publishing in a World with AI. Brevity Blog. |
| ChatGPT, Writers, and Publishing: Should You Worry? Heidi Pitlor Editoral. |
Writing
| AI Needs Journalism More Than Journalism Needs AI. Computer History Museum. |
| How can newsrooms benefit from ChatGPT and other generative AI tools? Hacks/Hackers. |